A method and system for building indoor semantic maps and planning paths for mobile robots

Abstract

The present invention relates to the field of robotics, and in particular to a method and system for building and planning indoor semantic maps of mobile robots, the method comprising: a mobile robot receives natural language instructions, a large language model outputs a navigation target after understanding the instructions, and the mobile robot creates a two-dimensional occupancy grid map within the current field of view according to the data of a depth camera; a target detection model is applied to perform target detection on the RGB image obtained by the depth camera to generate an object list; a large language model is applied to read the object position information and add regional attributes to the object list; a semantic topological map is created based on the two-dimensional occupancy grid map and the object position information with regional attributes; a navigation target is searched according to the semantic topological map, a path is planned, and a target navigation task is completed. The present invention constructs a semantic topological map of an indoor environment, realizes a semantic navigation task based on natural language instructions, and greatly improves the efficiency of large-scale indoor navigation in a complex space.

Description

Technical Field

[0001] The present invention relates to the field of robotics technology, and in particular to a method and system for indoor semantic map construction and path planning of a mobile robot. Background Art

[0002] In today's society, indoor navigation has become an indispensable part of daily life. Robots need to complete complex navigation tasks in a variety of scenarios such as shopping malls, hospitals, airports, and even home rooms. Research on indoor visual language navigation assisted by large language models (LLMs) helps improve the accessibility of indoor environments and is beneficial for helping the elderly, the blind, and other groups with special needs to complete indoor activities. Research on LLMs represented by ChatGPT provides strong support for the rapid development of embodied intelligence. Deploying LLMs on robots can control robots through natural language to complete tasks such as target navigation.

[0003] Although path planning for autonomous mobile robots has been widely studied, it is still a challenge for robots to achieve efficient and accurate navigation in complex and large-scale indoor environments. The main reasons are the lack of semantic information about the robot's mobile space and the lack of connection between the starting position and the navigation target. A semantic map is a map that describes different locations in the environment. Each location is accompanied by semantic information. The semantic information can include the type of object, the location of landmarks, etc. The purpose is to provide robots with a richer understanding of the environment. In recent years, semantic maps using machine learning neural networks require manual annotation. The method has poor universality and requires manual annotation, which is time-consuming and labor-intensive. Using pre-trained models to identify objects can complete the target detection task, thereby automatically adding semantic information to the raster map.

[0004] A topological map represents key locations in an environment and the connections between them. It is usually in the form of a graph, where nodes represent locations and edges represent connections. Its main purpose is to provide navigation paths for robots, allowing them to move through the connections between locations without the need for detailed geometric information. Traditional topological maps require manual addition of connections between key points, while the use of a large language model can read the list information generated by target detection and generate corresponding connections between each key point, which helps mobile robots plan paths when completing natural language instruction tasks. Mobile robots add semantic information and connections between key points on the grid map while moving, which is closer to the way humans navigate. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a method and system for indoor semantic map construction and path planning of a mobile robot.

[0006] The technical solution of the present invention is: a method for building an indoor semantic map and planning a path for a mobile robot, comprising:

[0007] Step S1: The mobile robot receives the user's natural language instructions, and uses the large language model to parse and reason the natural language instructions to obtain the navigation target;

[0008] Step S2: The mobile robot rotates in situ in the current mapping area, obtains a depth image of its surrounding environment through the depth camera carried by the mobile robot, and converts the depth information into laser information to create a two-dimensional occupancy grid map; at the same time, the target detection network is used to perform target detection on the RGB image of the surrounding environment obtained by the depth camera, generate an object information list, and use the depth image to calculate the coordinate information of the object in space, and convert it to obtain the corresponding position of the object in the two-dimensional occupancy grid map;

[0009] Step S3: Control the mobile robot to move to the next mapping area and go to step S2 until a complete two-dimensional occupancy grid map of the room is created;

[0010] Step S4: inputting the object information list into the large language model, adding regional attributes to the object information list; creating a semantic topological map based on the two-dimensional occupancy grid map and the object position information with regional attributes;

[0011] Step S5: determining whether the navigation target exists in the semantic topology map; if so, completing path planning in the semantic topology map according to the semantic topology map information, and navigating the mobile robot to the location of the navigation target;

[0012] Step S6: If it does not exist, through interaction with the user, use the large language model to infer the possible position of the navigation target in the semantic topology map, and continuously update the two-dimensional occupancy grid map and the semantic topology map during the process of the mobile robot navigating to the possible position until the navigation target is found, and go to step S5 to navigate the mobile robot to the position of the navigation target.

[0013] Compared with the prior art, the present invention has the following advantages:

[0014] 1. The present invention discloses a method for building an indoor semantic map and planning a path for a mobile robot. The method uses a target detection model to perform real-time target detection during the movement of the mobile robot, outputs a target detection list, obtains the location information of the object, and uses the confidence result of the target detection object to control the mobile robot to stop moving.

[0015] 2. The semantic navigation task based on natural language instructions in the present invention greatly improves the efficiency of large-scale indoor navigation in complex space. The large language model reads the existing target detection object list information, adds corresponding regional attributes to the object information, and creates a semantic topological map based on the two-dimensional occupancy grid map and the object position information with regional attributes. The information on whether the areas are connected and the object position information in the area provide a reference for the navigation planning of the mobile robot.

[0016] 3. In a complex indoor navigation environment, the mobile robot first calls the large language model to understand the natural language instructions and outputs the navigation target. For navigation tasks where the navigation target is in the semantic topological map, the mobile robot plans a minimum cost node sequence between the starting point and the navigation target, and the sub-paths between the nodes are completed with a common path planning algorithm in the original map to complete local path planning, and finally the sub-paths are connected together to obtain the final path. For navigation tasks where the navigation target is not in the semantic topological map, the large language model is used to infer the possible position of the navigation target in the semantic topological map through interaction with the user, and the two-dimensional occupancy grid map and the semantic topological map are continuously updated during the process of the mobile robot navigating to the possible position until the navigation target is found, and then the large language model continues to use the semantic topological map for planning until the navigation is completed by reasoning. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of a method for building an indoor semantic map and planning a path for a mobile robot in an embodiment of the present invention;

[0018] Figure 2 A schematic diagram of obtaining pixel coordinates for target detection in an embodiment of the present invention;

[0019] Figure 3 A schematic diagram of corresponding positions of pixel coordinates in a two-dimensional grid map in an embodiment of the present invention;

[0020] Figure 4 The object location information with regional attributes output by the large language model in the embodiment of the present invention;

[0021] Figure 5 This is a schematic diagram of a semantic topology map in an embodiment of the present invention;

[0022] Figure 6 This is a structural block diagram of a mobile robot indoor semantic map construction and path planning system in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The present invention provides a method for indoor semantic map construction and path planning of a mobile robot, which constructs a semantic topological map of an indoor environment and enables the mobile robot to complete navigation tasks of natural language instructions without obtaining a map in advance.

[0024] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below through specific implementations and in conjunction with the accompanying drawings.

[0025] Embodiment 1

[0026] like Figure 1 As shown, an embodiment of the present invention provides a method for building an indoor semantic map and planning a path for a mobile robot, comprising the following steps:

[0027] Step S1: The mobile robot receives the user's natural language instructions, and uses the large language model to parse and reason the natural language instructions to obtain the navigation target;

[0028] Step S2: The mobile robot rotates in situ in the current mapping area, and obtains a depth image of its surrounding environment through the depth camera carried by the mobile robot, converts the depth information into laser information to create a two-dimensional occupancy grid map; at the same time, the target detection network is used to perform target detection on the RGB image of the surrounding environment obtained by the depth camera, generates an object information list, and uses the depth image to calculate the coordinate information of the object in space, and converts it to obtain the corresponding position of the object in the two-dimensional occupancy grid map;

[0029] Step S3: Control the mobile robot to move to the next mapping area and go to step S2 until a complete two-dimensional occupancy grid map of the room is created;

[0030] Step S4: input the object information list into the large language model, add regional attributes to the object information list; create a semantic topological map based on the two-dimensional occupancy grid map and the object position information with regional attributes;

[0031] Step S5: Determine whether the navigation target exists in the semantic topology map; if so, complete path planning in the semantic topology map according to the semantic topology map information, and navigate the mobile robot to the location of the navigation target;

[0032] Step S6: If it does not exist, through interaction with the user, use the large language model to infer the possible position of the navigation target in the semantic topology map, and continuously update the two-dimensional occupancy grid map and the semantic topology map during the process of the mobile robot navigating to the possible position until the navigation target is found, and then go to step S5 to navigate the mobile robot to the position of the navigation target.

[0033] In one embodiment, the above step S1: the mobile robot receives the user's natural language instruction, and uses the large language model to parse and reason the natural language instruction to obtain the navigation target; specifically includes:

[0034] The mobile robot is in an indoor environment without a pre-given map, such as a bedroom. First, the mobile robot receives a natural language instruction from the user, such as "please move to the oven position". The mobile robot uses the API function parse_navigation_instruction (natural_language_instruction) of the large language model to parse the instruction and obtain the navigation target: oven.

[0035] In one embodiment, the above step S2: the mobile robot rotates in situ in the current mapping area, obtains a depth image of its surrounding environment through a depth camera carried by the mobile robot, and converts the depth information into laser information to create a two-dimensional occupancy grid map; at the same time, the target detection network is used to perform target detection on the RGB image of the surrounding environment obtained by the depth camera, generate an object information list, and use the depth image to calculate the coordinate information of the object in space, and convert it to obtain the corresponding position of the object in the two-dimensional occupancy grid map; specifically includes:

[0036] Step S21: Initialize the two-dimensional occupancy grid map and create a two-dimensional array, each element of which represents a grid in the two-dimensional occupancy grid map, and the grid is used to represent the state of the environment: occupied or free space;

[0037] The mobile robot in the embodiment of the present invention is equipped with an inertial measurement unit IMU and a RealSense depth camera. The mobile robot first rotates in place, and the depth camera obtains the current depth image and RGB image. Then, the three-dimensional image information released by the depth camera is converted into two-dimensional laser information. The two-dimensional laser information is combined with the odometer information to create the current two-dimensional occupancy grid map.

[0038] Step S22: Map each pixel of the depth image to a corresponding position of the two-dimensional occupancy grid map, convert the depth value into a grid state using a threshold, regard a depth less than the threshold as an obstacle, and regard a depth greater than the threshold as free space;

[0039] Step S23: During the navigation process of the mobile robot, new depth information is continuously acquired and the corresponding grid state is updated, thereby updating the two-dimensional occupancy grid map in real time;

[0040] Step S24: Collect and annotate data on the indoor environment, use the target detection model to detect indoor targets, store the target detection results in the form of a dictionary, the dictionary contains the name of the object and the pixel position information, calculate the two-dimensional position of the object through the depth information, and convert the pixel position information into the corresponding position in the two-dimensional grid map;

[0041] The embodiment of the present invention uses the target detection model YOLOv8 to perform timed target detection on the RGB image obtained by the current depth camera, limits the detection frequency, obtains the name of the object with a low repetition rate and its pixel coordinate information in the image, and uses the depth image to calculate the coordinates of the object in space, which are projected onto the map to obtain the position information of the object in the plane.

[0042] like Figure 2 The pixel coordinates obtained by target detection are shown. Figure 3 For the general Figure 2 The pixel coordinates are converted to the corresponding positions in the two-dimensional grid map.

[0043] In one embodiment, the above step S3: controlling the mobile robot to move to the next mapping area, and going to step S2 until a complete two-dimensional occupancy grid map of the room is created;

[0044] After the mobile robot obtains the two-dimensional occupancy grid map of the current mapping area through step S2, it moves to the next mapping area and repeats step S2 to continue constructing the two-dimensional occupancy grid map until a complete two-dimensional occupancy grid map of the bedroom is constructed.

[0045] In one embodiment, the above step S4: inputting the object information list into the large language model, adding regional attributes to the object list; creating a semantic topological map based on the two-dimensional occupancy grid map and the object position information with regional attributes, specifically includes:

[0046] Step S41: preprocessing the two-dimensional occupancy grid map to filter out its white noise to obtain an original map;

[0047] Step S42: using the large language model to add regional attributes to the object;

[0048] Step S43: using the common sense and prompts of the large language model, the connectivity between the regions is determined by the region name and the object information in the region, such as corridors or doors;

[0049] Step S44: Save the location information of objects with regional attributes as a structure, take a region as a node, and the connectivity relationship between regions is represented by 1 and 0 as weights to construct a weighted undirected network. Superimpose the original map and the weighted undirected network to obtain a semantic topological map, and continuously update the semantic topological map as the mobile robot moves.

[0050] The present invention designs a function to build a semantic topology map, calls the big language model API function build_semantic_topology_map(object_information_list), reads the above object location information file, and through prompts, the big language model uses common sense to add regional attributes to the existing object location information according to the function and attribute of the object, and outputs information such as Figure 4 As shown. Figure 5 As shown in the figure, the original map (left) and the weighted undirected network (right) are superimposed to obtain a semantic topological map: the weighted undirected network regards regions as nodes, and the connectivity between two regions as edges between two nodes, with edge weights of 1 or 0. Because the region contains the names and codes of objects in the region and the location information in the map coordinate system, the names and lists of object names and location information contained in the region are encapsulated into a structure, and the Vex[vexsnum] structure array is used to store the basic information of all locations in the map, where vexsnum is the number of regions.

[0051] After the mobile robot completes the creation of the entire two-dimensional occupancy grid map in step S3, the large language model can determine that the area is a bedroom through the object list information such as bedroom doors, table lamps, curtains, and bedside tables. The large language model outputs object location information with regional attributes such as Figure 4 As shown in the figure, as the mobile robot moves, the object location information is continuously updated, and the large language model adds new object information and regional attributes to the semantic topology map in real time.

[0052] In one embodiment, the above step S5: determining whether the navigation target exists in the semantic topology map, and if so, completing path planning in the semantic topology map according to the semantic topology map information to navigate the mobile robot to the location of the navigation target, specifically includes:

[0053] Step S51: determining whether there is an object name of the navigation target in the semantic topological map;

[0054] Step S52: If it exists, the mobile robot plans a minimum cost node sequence between the starting point and the navigation target according to the location of the navigation target in the current semantic topology map, and the sub-paths between the nodes are planned by a conventional path planning algorithm, such as a Planning is performed and finally all sub-paths are connected together to obtain the final path; and the semantic topology map and the two-dimensional occupancy grid map are updated in real time during the navigation process until the mobile robot reaches the location of the navigation target.

[0055] In one embodiment, the above step S6: if it does not exist, by interacting with the user, the large language model is used to infer the possible position of the navigation target in the semantic topology map, and the two-dimensional occupancy grid map and the semantic topology map are continuously updated during the process of the mobile robot navigating to the possible position until the navigation target is found, and the mobile robot is navigated to the position of the navigation target in step S5, which specifically includes:

[0056] In this step, the embodiment of the present invention uses the large language model API function explore_to_target (navigation_target, semantic_topology_map) to read the current semantic topology map, and uses the large language model to infer the possible position of the navigation target in the semantic topology map. For example, in navigation, the mobile robot is currently located in the bedroom. Because of the fine-grained problem of step S3 map construction, there may be a situation where the navigation target "oven" is not included in the current semantic topology map. The large language model infers possible position 1 based on the existing semantic topology map information, such as: bedroom door, and completes navigation from the starting point to possible position 1 according to the navigation steps within the visible range. The semantic topology map is updated in real time during movement. After reaching possible position 1, possible position 1 is used as the starting point. The large language model determines whether the navigation target exists according to the current semantic topology map. If it exists, it goes to step S5; if it does not exist, it continues to infer possible position 2, and navigates from possible position 1 to possible position 2. Then, possible position 2 is used as the starting point to continue to determine whether the navigation target exists in the current semantic topology map until the navigation target exists in the current semantic topology map. In this process, the two-dimensional occupancy grid map and the semantic topology map are updated in real time.

[0057] Embodiment 2

[0058] like Figure 6 As shown, an embodiment of the present invention provides a mobile robot indoor semantic map construction and path planning system, including the following modules:

[0059] Obtaining a navigation target module 71, for inputting a natural language instruction to the mobile robot, parsing and reasoning the natural language instruction using a large language model, and obtaining a navigation target;

[0060] The module 72 for generating a two-dimensional occupancy grid map is used for the mobile robot to rotate in situ in the current mapping area, and to obtain a depth image of its surrounding environment through a depth camera carried by the mobile robot, and to convert the depth information into laser information to create a two-dimensional occupancy grid map; at the same time, the target detection network is used to perform target detection on the RGB image of the surrounding environment obtained by the depth camera, to generate an object information list, and to calculate the coordinate information of the object in space using the depth image, and to convert the corresponding position of the object in the two-dimensional occupancy grid map;

[0061] The module 73 for constructing a complete indoor two-dimensional occupancy grid map is used to control the mobile robot to move to the next mapping area and switch to the module for generating a two-dimensional occupancy grid map until a complete indoor two-dimensional occupancy grid map is created;

[0062] A semantic topology map building module 74 is used to input the object information list into the large language model, add regional attributes to the object information list; create a semantic topology map based on the two-dimensional occupancy grid map and the object position information with regional attributes;

[0063] The navigation module 75 is used to determine whether the navigation target exists in the semantic topology map; if so, in the semantic topology map, the path planning is completed according to the semantic topology map information to navigate the mobile robot to the location of the navigation target;

[0064] The navigation target search module 76, if the navigation target does not exist in the semantic topology map, interacts with the user and uses the large language model to infer the possible position of the navigation target in the semantic topology map. During the process of the mobile robot navigating to the possible position, the two-dimensional occupancy grid map and the semantic topology map are continuously updated until the navigation target is found, and then the navigation module is transferred to navigate the mobile robot to the position of the navigation target.

[0065] The above embodiments are provided only for the purpose of describing the present invention, and are not intended to limit the scope of the present invention. The scope of the present invention is defined by the appended claims. Various equivalent substitutions and modifications made without departing from the spirit and principles of the present invention should all be included within the scope of the present invention.

Claims

1. A method for constructing indoor semantic maps and path planning for a mobile robot, characterized in that: include: Step S1: The mobile robot receives the user's natural language instructions, and uses the large language model to parse and reason the natural language instructions to obtain the navigation target; Step S2: The mobile robot rotates in situ in the current mapping area, and obtains a depth image of its surrounding environment through the depth camera carried by the mobile robot, converts the depth information into laser information to create a two-dimensional occupancy grid map; at the same time, the target detection network is used to perform target detection on the RGB image of the surrounding environment obtained by the depth camera, generates an object information list, and uses the depth image to calculate the coordinate information of the object in space, and converts the corresponding position of the object in the two-dimensional occupancy grid map, specifically including: Step S21: Initialize the two-dimensional occupancy grid map and create a two-dimensional array, each element of which represents a grid in the two-dimensional occupancy grid map, and the grid is used to represent the state of the environment: occupied or free space; Step S22: Map each pixel of the depth image to a corresponding position of the two-dimensional occupancy grid map, convert the depth value into a grid state using a threshold, regard a depth less than the threshold as an obstacle, and regard a depth greater than the threshold as free space; Step S23: During the navigation process of the mobile robot, new depth information is continuously acquired and the corresponding grid state is updated, thereby updating the two-dimensional occupancy grid map in real time; Step S24: Detect indoor objects using the object detection model. The object detection results are stored in the form of a dictionary, wherein the dictionary contains the name of the object and pixel position information. The two-dimensional position of the object is calculated using the depth information, and the pixel position information is converted into a corresponding position in the two-dimensional occupancy grid map. Step S3: Control the mobile robot to move to the next mapping area and go to step S2 until a complete two-dimensional occupancy grid map of the room is created; Step S4: inputting the object information list into the large language model, adding regional attributes to the object information list; creating a semantic topological map based on the two-dimensional occupancy grid map and the object position information with regional attributes, specifically including: Step S41: preprocessing the two-dimensional occupancy grid map to filter out white noise and obtain an original map; Step S42: using the large language model to add regional attributes to the object; Step S43: using the common sense and prompts of the large language model, the connectivity between the regions is determined by the region name and the object information in the region: corridor or door; Step S44: The location information of the object with the region attribute is saved as a structure, and a region is used as a node. The connectivity relationship between regions is represented by 1 and 0 as weights, and a weighted undirected network is constructed. The original map and the weighted undirected network are superimposed to obtain a semantic topological map, and the semantic topological map is continuously updated as the mobile robot moves; Step S5: determining whether the navigation target exists in the semantic topology map; if so, completing path planning in the semantic topology map according to the semantic topology map information, and navigating the mobile robot to the location of the navigation target; Step S6: If it does not exist, through interaction with the user, use the large language model to infer the possible position of the navigation target in the semantic topology map, and continuously update the two-dimensional occupancy grid map and the semantic topology map during the process of the mobile robot navigating to the possible position until the navigation target is found, and go to step S5 to navigate the mobile robot to the position of the navigation target.

2. The method for indoor semantic map construction and path planning of a mobile robot according to claim 1, characterized in that: The step S5: determining whether the navigation target exists in the semantic topology map, and if so, completing path planning in the semantic topology map according to the semantic topology map information, and navigating the mobile robot to the location of the navigation target, specifically includes: Step S51: determining whether the object name of the navigation target exists in the semantic topology map; Step S52: If it exists, according to the position of the navigation target in the current semantic topological map, the mobile robot plans a minimum cost node sequence between the starting point and the navigation target, and the sub-paths between the nodes are planned by a conventional path planning algorithm, and finally all sub-paths are connected together to obtain the final path; and the semantic topological map and the two-dimensional occupancy grid map are updated in real time during the navigation process until the mobile robot reaches the position of the navigation target.

3. The method for indoor semantic map construction and path planning of a mobile robot according to claim 2, characterized in that: The step S6: if it does not exist, the possible position of the navigation target in the semantic topology map is inferred by using the large language model through interaction with the user, and the two-dimensional occupancy grid map and the semantic topology map are continuously updated during the process of the mobile robot navigating to the possible position until the navigation target is found, and the mobile robot is navigated to the position of the navigation target in step S5, which specifically includes: Step S61: determining whether the object name of the navigation target exists in the semantic topological map; Step S62: If it does not exist, by interacting with the user and using a large language model to read the current semantic topology map, the possible position of the navigation target in the semantic topology map is inferred, and the mobile robot uses the possible position as a temporary navigation target, and plans a minimum cost node sequence between the starting point and the temporary navigation target. The sub-paths between the nodes are planned using a conventional path planning algorithm, and finally all sub-paths are connected together, the mobile robot is navigated to the possible position, the semantic topology map is updated, and the possible position is used as the starting point, and then go to step S51 to continue judging; otherwise, go to step S62.

4. A mobile robot indoor semantic map construction and path planning system, characterized in that: Includes the following modules: The navigation target acquisition module is used to input natural language instructions to the mobile robot, and use the large language model to parse and reason about the natural language instructions to obtain the navigation target; The module for generating a two-dimensional occupancy grid map is used for the mobile robot to rotate in situ in the current mapping area, and the depth image of its surrounding environment obtained by the depth camera carried by the mobile robot is used to convert the depth information into laser information to create a two-dimensional occupancy grid map; at the same time, the target detection network is used to perform target detection on the RGB image of the surrounding environment obtained by the depth camera, generate an object information list, and use the depth image to calculate the coordinate information of the object in space, and convert it to obtain the corresponding position of the object in the two-dimensional occupancy grid map, which specifically includes: Step S21: Initialize the two-dimensional occupancy grid map and create a two-dimensional array, each element of which represents a grid in the two-dimensional occupancy grid map, and the grid is used to represent the state of the environment: occupied or free space; Step S22: Map each pixel of the depth image to a corresponding position of the two-dimensional occupancy grid map, convert the depth value into a grid state using a threshold, regard a depth less than the threshold as an obstacle, and regard a depth greater than the threshold as free space; Step S23: During the navigation process of the mobile robot, new depth information is continuously acquired and the corresponding grid state is updated, thereby updating the two-dimensional occupancy grid map in real time; Step S24: Detect indoor objects using the object detection model. The object detection results are stored in the form of a dictionary, wherein the dictionary contains the name of the object and pixel position information. The two-dimensional position of the object is calculated using the depth information, and the pixel position information is converted into a corresponding position in the two-dimensional occupancy grid map. A module for constructing a complete indoor two-dimensional occupancy grid map is used to control the mobile robot to move to the next mapping area and switch to the module for generating a two-dimensional occupancy grid map until a complete indoor two-dimensional occupancy grid map is created; Constructing a semantic topological map module, for inputting the object information list into a large language model, and adding regional attributes to the object information list; creating a semantic topological map based on the two-dimensional occupancy grid map and the object location information with regional attributes, specifically including: Step S41: preprocessing the two-dimensional occupancy grid map to filter out white noise and obtain an original map; Step S42: using the large language model to add regional attributes to the object; Step S43: using the common sense and prompts of the large language model, the connectivity between the regions is determined by the region name and the object information in the region: corridor or door; Step S44: The location information of the object with the region attribute is saved as a structure, and a region is used as a node. The connectivity relationship between regions is represented by 1 and 0 as weights, and a weighted undirected network is constructed. The original map and the weighted undirected network are superimposed to obtain a semantic topological map, and the semantic topological map is continuously updated as the mobile robot moves; A navigation module is used to determine whether the navigation target exists in the semantic topology map; if so, in the semantic topology map, path planning is completed according to the semantic topology map information to navigate the mobile robot to the location of the navigation target; A navigation target search module, if the navigation target does not exist in the semantic topological map, interacts with the user and utilizes a large language model to infer the possible position of the navigation target in the semantic topological map. During the process of the mobile robot navigating to the possible position, the two-dimensional occupancy grid map and the semantic topological map are continuously updated until the navigation target is found, and the navigation module is turned to navigate the mobile robot to the position of the navigation target.

Images